ray is first initialized explicitly to set maximum resources limits, becuase it does not respect nextflow and cluster limits otherwise.
But doing so if num_gpus is left to None, default for ray.init see here -> docs the GPUs are not automaticly detected and used. In theory it should detect them all and use them all, but is not doing so.
ray is first initialized explicitly to set maximum resources limits, becuase it does not respect nextflow and cluster limits otherwise.
But doing so if num_gpus is left to None, default for ray.init see here -> docs the GPUs are not automaticly detected and used. In theory it should detect them all and use them all, but is not doing so.